Modeling the GOP Nomination: Before Indiana

 

As described in sections 3.2 and 3.3 of the new PDF, the prediction model performed well last week. The median predicted delegates for Trump-Cruz-Kasich were 108-4-6, which compares well to the actual final delegate totals of 111-2-5.

The delegate model converted the actual voting shares in the states to actual delegates almost perfectly (only off by one in Rhode Island), and the model’s 80-percent confidence intervals for voting shares and delegates happened to contain the true value in all cases.

Chapter 4 shows updated model parameters intended to predict going forward. Table 4.2 shows probabilities of winning Indiana: Trump about 90 percent, Cruz about nine percent, and Kasich less than one percent. The vote-share 80-percent confidence intervals for today’s primary and Nebraska’s next week are as follows:

Trump: 41–66 percent (Indiana) and 31–57 percent (Nebraska)

Cruz: 17–41 (Indiana) and 27–55 percent (Nebraska)

Kasich: 10–26 (Indiana) and 8–23 percent (Nebraska)

Adding everything together, Trump’s probability of winning on the first ballot is now 94 percent according to the model (Table 4.5). That looks bleak for Cruz but, if Cruz wins Indiana, Trump’s probability of winning on the first ballot falls immediately to about 70 percent (Table 4.9). Indiana does indeed look rather like “do or die” for Cruz.

If Cruz wins this evening, every future state looks better for him, especially California. For example, the Nebraska 80-percent confidence intervals would become:

Trump: 21–42 percent

Cruz: 44–68 percent

Kasich: 7–21 percent

Under this scenario, the probability of Cruz winning Nebraska and California rises from 48 and 7 percent (respectively) to 95 percent and 22 percent. Overall, the expected net delegate gain for Cruz relative to Trump if Cruz wins Indiana is about 270 delegates. This dynamic is a result of our correlation assumptions.

So provided Cruz wins Indiana, he would still have decent prospects of keeping Trump from 1237.

Published in General
Like this post? Want to comment? Join Ricochet’s community of conservatives and be part of the conversation. Join Ricochet for Free.

There are 6 comments.

Become a member to join the conversation. Or sign in if you're already a member.
  1. Rodin Member
    Rodin
    @Rodin

    Yes, looks like Indiana is the pivot point. Cruz is probably “dead man walking” if Trump wins; Trump’s inevitability narrative is dealt a strong blow if Cruz wins.

    • #1
  2. Hoyacon Member
    Hoyacon
    @Hoyacon

    Thank you, Chuck.  Really like your stuff.

    • #2
  3. Matt Bartle Member
    Matt Bartle
    @MattBartle

    Very interesting. I think Trump is going to do this, even though I’ve been thinking that one of these days he was going to fade.

    Have you applied this model to the general election in November??

    • #3
  4. Marion Evans Inactive
    Marion Evans
    @MarionEvans

    “It was my understanding that there would be no math.”

    • #4
  5. John Wilson Member
    John Wilson
    @

    Yeah, I don’t think it requires a lot of statistical analysis or modeling to know that Indiana is the ballgame. Simple arithmetic and a look at the polls tells you all you need to know.

    That said, Trump is going to win very decisively tonight, and I for one will not be welcoming our new alt-right, protectionist, isolationist, statist, incoherent, dishonest, corrupt, populist overlords.

    • #5
  6. Chuck Walla Member
    Chuck Walla
    @ChuckWalla

    John Wilson:Yeah, I don’t think it requires a lot of statistical analysis or modeling to know that Indiana is the ballgame. Simple arithmetic and a look at the polls tells you all you need to know.

    The point of this model is not merely to repeat what others already suspect about the Indiana primary, though I did want to measure this.  The point of the model is more to test and demonstrate new modeling techniques for broader application.  My posts are not just about the primaries themselves, but also about testing a model in real time.

    My focus for the primaries are the delegates more than voting shares.  A campaign, or a PAC, has limited resources, so choices must be made.  The best use of this particular model is probably to decide what spending or campaigning choice gives you the most expected delegate bang for the buck or minute.

    For example, it is possible to guess that a marginal dollar spent in Oregon is not worth as much as spending elsewhere (due to the proportional awarding of delegates in Oregon, both statewide and by congressional district), but how sure are you given the intricacies of the delegate-allocation rules?  My model spends 98% of its run-time modeling congressional-district outcomes, and I am pleased with how well it has done so far.

    By the way, this modeling technique (possibly unrecognizable to well-trained frequentists) is in line with Hubbard’s book, “How to Measure Anything.”  It is also consistent with the methods and philosophy of E.T. Jaynes’s use of probability theory as logic.  In brief, partial information is often best modeled with random variables, employing whatever information you have. My model contains a few new tools, not commonly used, that I wanted to try out before going to a more demanding application.

    • #6
Become a member to join the conversation. Or sign in if you're already a member.